The invention relates to a near-field sound source positioning method based on partial least squares regression and can effectively solve multiple correlation problems between variables. The method ischaracterized in that L sets of receiving data generated by K narrowband, non-Gaussian and stationary near-field sound source signals in a training interval are received by a uniform linear array, after each set of receiving data is subjected to covariance to obtain a corresponding covariance matrix, upper triangular elements of the data covariance matrix are extracted, standardization processingis performed, a training sample source set is subjected to standardization processing, the number of extracted components is determined based on the cross validity, and thereby a satisfactory estimation model is obtained; test data is estimated through utilizing the trained near-field source partial least squares regression model, the angle and the distance of a test source are estimated; the components extracted by partial least squares regression are not only a good summary of the information in an independent variable system, but also explains dependent variables well, and eliminates noiseinterference in the system, and the predicted angle and the distance are highly accurate.